|
|
@@ -0,0 +1,403 @@
|
|
|
+/* This file is part of DarkFi (https://dark.fi)
|
|
|
+ *
|
|
|
+ * Copyright (C) 2020-2023 Dyne.org foundation
|
|
|
+ *
|
|
|
+ * This program is free software: you can redistribute it and/or modify
|
|
|
+ * it under the terms of the GNU Affero General Public License as
|
|
|
+ * published by the Free Software Foundation, either version 3 of the
|
|
|
+ * License, or (at your option) any later version.
|
|
|
+ *
|
|
|
+ * This program is distributed in the hope that it will be useful,
|
|
|
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
+ * GNU Affero General Public License for more details.
|
|
|
+ *
|
|
|
+ * You should have received a copy of the GNU Affero General Public License
|
|
|
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
|
|
|
+ */
|
|
|
+
|
|
|
+//! Chunk-based file storage implementation.
|
|
|
+//! This is a building block for a DHT or something similar.
|
|
|
+//!
|
|
|
+//! The API supports file insertion and retrieval. There is intentionally no
|
|
|
+//! `remove` support. File removal should be handled externally, and then it
|
|
|
+//! is only required to run `garbage_collect()` to clean things up.
|
|
|
+//!
|
|
|
+//! The filesystem hierarchy stores two directories: `files` and `chunks`.
|
|
|
+//! `chunks` store [`MAX_CHUNK_SIZE`] files, where the filename is a BLAKE3
|
|
|
+//! hash of the chunk's contents.
|
|
|
+//! `files` store metadata about a full file, which can be retrieved by
|
|
|
+//! concatenating the chunks in order. The filename of a file in `files`
|
|
|
+//! is the BLAKE3 hash of hashed chunks in the correct order.
|
|
|
+//!
|
|
|
+//! It might look like the following:
|
|
|
+//! ```
|
|
|
+//! /files/7d4c0d5539057c8f9b60d32b423964beb38ecd8ea1ab203c0207990cbf0cad22
|
|
|
+//! /files/...
|
|
|
+//! /chunks/9d7abc2efa52b8be63ff82b756edb6822e09aa40fc587aba977185a5bb449c19
|
|
|
+//! /chunks/fc432e087d16d8788e87640511e627be34a4a50533f1e5ed3e1370645a0266b8
|
|
|
+//! /chunks/...
|
|
|
+//! ```
|
|
|
+//!
|
|
|
+//! In the above example, contents of `7d4c0d5539057c8f9b60d32b423964beb38ecd8ea1ab203c0207990cbf0cad22`
|
|
|
+//! may be:
|
|
|
+//! ```
|
|
|
+//! 9d7abc2efa52b8be63ff82b756edb6822e09aa40fc587aba977185a5bb449c19
|
|
|
+//! fc432e087d16d8788e87640511e627be34a4a50533f1e5ed3e1370645a0266b8
|
|
|
+//! ```
|
|
|
+//!
|
|
|
+//! This means, in order to retrieve `7d4c0d5539057c8f9b60d32b423964beb38ecd8ea1ab203c0207990cbf0cad22`,
|
|
|
+//! we need to concatenate the files under `/chunks` whose filenames are the
|
|
|
+//! hashes found above. The contents of the files in `/chunks` are arbitrary
|
|
|
+//! data, and by concatenating them we can retrieve the original file.
|
|
|
+//!
|
|
|
+//! It is important to note that multiple files can use the same chunks.
|
|
|
+//! This is some kind of naive deduplication, so we actually don't consider
|
|
|
+//! chunks to be specific to a single file and therefore when we do garbage
|
|
|
+//! collection, we keep chunks and files independent of each other.
|
|
|
+
|
|
|
+use std::collections::HashSet;
|
|
|
+
|
|
|
+use async_std::{
|
|
|
+ fs::{self, File, OpenOptions},
|
|
|
+ io::{prelude::*, BufReader, Cursor, SeekFrom},
|
|
|
+ path::PathBuf,
|
|
|
+ stream::StreamExt,
|
|
|
+};
|
|
|
+use log::{debug, info, warn};
|
|
|
+
|
|
|
+use crate::{Error, Result};
|
|
|
+
|
|
|
+/// Defined maximum size of a stored chunk (256 KiB)
|
|
|
+pub const MAX_CHUNK_SIZE: usize = 262_144;
|
|
|
+
|
|
|
+/// Path prefix where file metadata is stored
|
|
|
+const FILES_PATH: &str = "files";
|
|
|
+/// Path prefix where file chunks are stored
|
|
|
+const CHUNKS_PATH: &str = "chunks";
|
|
|
+
|
|
|
+/// `ChunkedFile` is a representation of a file we're trying to
|
|
|
+/// retrieve from `Geode`. The tuple contains `blake3::Hash` of
|
|
|
+/// the file's chunks and an optional `PathBuf` which points to
|
|
|
+/// the filesystem where the chunk can be found. If `None`, it
|
|
|
+/// is to be assumed that the chunk is not available locally.
|
|
|
+#[derive(Clone)]
|
|
|
+pub struct ChunkedFile(Vec<(blake3::Hash, Option<PathBuf>)>);
|
|
|
+
|
|
|
+impl ChunkedFile {
|
|
|
+ fn new(hashes: &[blake3::Hash]) -> Self {
|
|
|
+ Self(hashes.iter().map(|x| (*x, None)).collect())
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Check whether we have all the chunks available locally.
|
|
|
+ pub fn is_complete(&self) -> bool {
|
|
|
+ !self.0.iter().any(|(_, p)| p.is_none())
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Return an iterator over the chunks and their paths.
|
|
|
+ pub fn iter(&self) -> core::slice::Iter<'_, (blake3::Hash, Option<PathBuf>)> {
|
|
|
+ self.0.iter()
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+/// Chunk-based file storage interface.
|
|
|
+pub struct Geode {
|
|
|
+ /// Path to the filesystem directory where file metadata is stored
|
|
|
+ files_path: PathBuf,
|
|
|
+ /// Path to the filesystem directory where file chunks are stored
|
|
|
+ chunks_path: PathBuf,
|
|
|
+}
|
|
|
+
|
|
|
+impl Geode {
|
|
|
+ /// Instantiate a new [`Geode`] object.
|
|
|
+ /// `base_path` defines the root directory where Geode will store its
|
|
|
+ /// file metadata and chunks.
|
|
|
+ pub async fn new(base_path: &PathBuf) -> Result<Self> {
|
|
|
+ let mut files_path: PathBuf = base_path.into();
|
|
|
+ let mut chunks_path: PathBuf = base_path.into();
|
|
|
+ files_path.push(FILES_PATH);
|
|
|
+ chunks_path.push(CHUNKS_PATH);
|
|
|
+
|
|
|
+ // Create necessary directory structure if needed
|
|
|
+ fs::create_dir_all(&files_path).await?;
|
|
|
+ fs::create_dir_all(&chunks_path).await?;
|
|
|
+
|
|
|
+ // Instantiate Self and perform initial garbage collection.
|
|
|
+ let self_ = Self { files_path, chunks_path };
|
|
|
+ self_.garbage_collect().await?;
|
|
|
+ Ok(self_)
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Attempt to read chunk hashes from a given file path and return
|
|
|
+ /// a `Vec` containing the hashes in order.
|
|
|
+ async fn read_metadata(path: &PathBuf) -> Result<Vec<blake3::Hash>> {
|
|
|
+ debug!(target: "geode::read_metadata()", "Reading chunks from {:?}", path);
|
|
|
+ let fd = File::open(path).await?;
|
|
|
+ let mut read_chunks = vec![];
|
|
|
+ let mut lines = BufReader::new(fd).lines();
|
|
|
+ while let Some(line) = lines.next().await {
|
|
|
+ let line = line?;
|
|
|
+ let chunk_hash = blake3::Hash::from_hex(line)?;
|
|
|
+ read_chunks.push(chunk_hash);
|
|
|
+ }
|
|
|
+
|
|
|
+ Ok(read_chunks)
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Perform garbage collection over the filesystem hierarchy.
|
|
|
+ /// Returns sets representing deleted files and deleted chunks, respectively.
|
|
|
+ pub async fn garbage_collect(&self) -> Result<(HashSet<blake3::Hash>, HashSet<blake3::Hash>)> {
|
|
|
+ info!(target: "geode::garbage_collect()", "[Geode] Performing garbage collection");
|
|
|
+ // We track corrupt files and chunks here.
|
|
|
+ let mut deleted_files = HashSet::new();
|
|
|
+ let mut deleted_chunks = HashSet::new();
|
|
|
+ let mut deleted_chunk_paths = HashSet::new();
|
|
|
+
|
|
|
+ // Scan through available chunks and check them for consistency.
|
|
|
+ let mut chunk_paths = fs::read_dir(&self.chunks_path).await?;
|
|
|
+ let mut buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ while let Some(chunk) = chunk_paths.next().await {
|
|
|
+ let Ok(entry) = chunk else { continue };
|
|
|
+ let chunk_path = entry.path();
|
|
|
+
|
|
|
+ // Skip if we're not a plain file
|
|
|
+ if !chunk_path.is_file().await {
|
|
|
+ continue
|
|
|
+ }
|
|
|
+
|
|
|
+ // Make sure that the filename is a BLAKE3 hash
|
|
|
+ let file_name = match chunk_path.file_name().and_then(|n| n.to_str()) {
|
|
|
+ Some(v) => v,
|
|
|
+ None => continue,
|
|
|
+ };
|
|
|
+ let chunk_hash = match blake3::Hash::from_hex(file_name) {
|
|
|
+ Ok(v) => v,
|
|
|
+ Err(_) => continue,
|
|
|
+ };
|
|
|
+
|
|
|
+ // If there is a problem with opening the file, remove it.
|
|
|
+ let Ok(mut chunk_fd) = File::open(&chunk_path).await else {
|
|
|
+ deleted_chunk_paths.insert(chunk_path);
|
|
|
+ deleted_chunks.insert(chunk_hash);
|
|
|
+ continue
|
|
|
+ };
|
|
|
+
|
|
|
+ // Perform consistency check
|
|
|
+ let Ok(bytes_read) = chunk_fd.read(&mut buf).await else {
|
|
|
+ deleted_chunk_paths.insert(chunk_path);
|
|
|
+ deleted_chunks.insert(chunk_hash);
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ continue
|
|
|
+ };
|
|
|
+
|
|
|
+ let chunk_slice = &buf[..bytes_read];
|
|
|
+ let hashed_chunk = blake3::hash(chunk_slice);
|
|
|
+
|
|
|
+ // If the hash doesn't match the filename, remove it.
|
|
|
+ if chunk_hash != hashed_chunk {
|
|
|
+ deleted_chunk_paths.insert(chunk_path);
|
|
|
+ deleted_chunks.insert(chunk_hash);
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ continue
|
|
|
+ }
|
|
|
+
|
|
|
+ // Seems legit.
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ }
|
|
|
+
|
|
|
+ for chunk_path in &deleted_chunk_paths {
|
|
|
+ if let Err(e) = fs::remove_file(chunk_path).await {
|
|
|
+ warn!(
|
|
|
+ target: "geode::garbage_collect()",
|
|
|
+ "[Geode] Garbage collect failed to remove corrupted chunk: {}", e,
|
|
|
+ );
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ // Perform health check over file metadata. For now we just ensure they
|
|
|
+ // have the correct format.
|
|
|
+ let mut file_paths = fs::read_dir(&self.files_path).await?;
|
|
|
+ while let Some(file) = file_paths.next().await {
|
|
|
+ let Ok(entry) = file else { continue };
|
|
|
+ let path = entry.path();
|
|
|
+
|
|
|
+ // Skip if we're not a plain file
|
|
|
+ if !path.is_file().await {
|
|
|
+ continue
|
|
|
+ }
|
|
|
+
|
|
|
+ // Make sure that the filename is a BLAKE3 hash
|
|
|
+ let file_name = match path.file_name().and_then(|n| n.to_str()) {
|
|
|
+ Some(v) => v,
|
|
|
+ None => continue,
|
|
|
+ };
|
|
|
+ let file_hash = match blake3::Hash::from_hex(file_name) {
|
|
|
+ Ok(v) => v,
|
|
|
+ Err(_) => continue,
|
|
|
+ };
|
|
|
+
|
|
|
+ // The filename is a BLAKE3 hash. It should contain a newline-separated
|
|
|
+ // list of chunks which represent the full file. If that is not the case
|
|
|
+ // we will consider it a corrupted file and delete it.
|
|
|
+ if Self::read_metadata(&path).await.is_err() {
|
|
|
+ if let Err(e) = fs::remove_file(path).await {
|
|
|
+ warn!(
|
|
|
+ target: "geode::garbage_collect()",
|
|
|
+ "[Geode] Garbage collect failed to remove corrupted file: {}", e,
|
|
|
+ );
|
|
|
+ }
|
|
|
+
|
|
|
+ deleted_files.insert(file_hash);
|
|
|
+ continue
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ info!(target: "geode::garbage_collect()", "[Geode] Garbage collection finished");
|
|
|
+ Ok((deleted_files, deleted_chunks))
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Insert a file into Geode. The function expects any kind of byte stream, which
|
|
|
+ /// can either be another file on the filesystem, a buffer, etc.
|
|
|
+ /// Returns a tuple of `(blake3::Hash, Vec<blake3::Hash>)` which represents the
|
|
|
+ /// file name, and the file's chunks, respectively.
|
|
|
+ pub async fn insert(
|
|
|
+ &self,
|
|
|
+ stream: impl AsRef<[u8]>,
|
|
|
+ ) -> Result<(blake3::Hash, Vec<blake3::Hash>)> {
|
|
|
+ info!(target: "geode::insert()", "[Geode] Inserting file...");
|
|
|
+ let mut file_hasher = blake3::Hasher::new();
|
|
|
+ let mut chunk_hashes = vec![];
|
|
|
+
|
|
|
+ let mut cursor = Cursor::new(&stream);
|
|
|
+ let mut buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+
|
|
|
+ while let Ok(bytes_read) = cursor.read(&mut buf).await {
|
|
|
+ if bytes_read == 0 {
|
|
|
+ break
|
|
|
+ }
|
|
|
+
|
|
|
+ let chunk_slice = &buf[..bytes_read];
|
|
|
+ let chunk_hash = blake3::hash(chunk_slice);
|
|
|
+ file_hasher.update(chunk_slice);
|
|
|
+ chunk_hashes.push(chunk_hash);
|
|
|
+
|
|
|
+ // Write the chunk to a file, if necessary. We first perform
|
|
|
+ // a consistency check and if things are fine, we don't have
|
|
|
+ // to perform a write, which is usually more expensive than
|
|
|
+ // reading from disk.
|
|
|
+ let mut chunk_path = self.chunks_path.clone();
|
|
|
+ chunk_path.push(chunk_hash.to_hex().as_str());
|
|
|
+ let mut chunk_fd =
|
|
|
+ OpenOptions::new().read(true).write(true).create(true).open(&chunk_path).await?;
|
|
|
+
|
|
|
+ let mut fs_buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ let fs_bytes_read = chunk_fd.read(&mut fs_buf).await?;
|
|
|
+ let fs_chunk_slice = &fs_buf[..fs_bytes_read];
|
|
|
+ let fs_chunk_hash = blake3::hash(fs_chunk_slice);
|
|
|
+
|
|
|
+ if fs_chunk_hash != chunk_hash {
|
|
|
+ debug!(
|
|
|
+ target: "geode::insert()",
|
|
|
+ "Existing chunk inconsistent or unavailable. Writing chunk to {:?}",
|
|
|
+ chunk_path,
|
|
|
+ );
|
|
|
+ // Here the chunk is broken, so we'll truncate and write the new one.
|
|
|
+ chunk_fd.set_len(0).await?;
|
|
|
+ chunk_fd.seek(SeekFrom::Start(0)).await?;
|
|
|
+ chunk_fd.write_all(chunk_slice).await?;
|
|
|
+ } else {
|
|
|
+ debug!(
|
|
|
+ target: "geode::insert()",
|
|
|
+ "Existing chunk consistent. Skipping write to {:?}",
|
|
|
+ chunk_path,
|
|
|
+ );
|
|
|
+ }
|
|
|
+
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ }
|
|
|
+
|
|
|
+ // This hash is the file's chunks hashed in order.
|
|
|
+ let file_hash = file_hasher.finalize();
|
|
|
+ let mut file_path = self.files_path.clone();
|
|
|
+ file_path.push(file_hash.to_hex().as_str());
|
|
|
+
|
|
|
+ // We always overwrite the metadata.
|
|
|
+ let mut file_fd = File::create(&file_path).await?;
|
|
|
+ for ch in &chunk_hashes {
|
|
|
+ file_fd.write(format!("{}\n", ch.to_hex().as_str()).as_bytes()).await?;
|
|
|
+ }
|
|
|
+
|
|
|
+ Ok((file_hash, chunk_hashes))
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Create and insert a single chunk into Geode given a stream.
|
|
|
+ /// Always overwrites any existing chunk. Returns the chunk hash once inserted.
|
|
|
+ pub async fn insert_chunk(&mut self, stream: impl AsRef<[u8]>) -> Result<blake3::Hash> {
|
|
|
+ info!(target: "geode::insert_chunk()", "[Geode] Inserting single chunk");
|
|
|
+
|
|
|
+ let mut cursor = Cursor::new(&stream);
|
|
|
+ let mut chunk = [0u8; MAX_CHUNK_SIZE];
|
|
|
+
|
|
|
+ let bytes_read = cursor.read(&mut chunk).await?;
|
|
|
+ let chunk_slice = &chunk[..bytes_read];
|
|
|
+ let chunk_hash = blake3::hash(chunk_slice);
|
|
|
+
|
|
|
+ let mut chunk_path = self.chunks_path.clone();
|
|
|
+ chunk_path.push(chunk_hash.to_hex().as_str());
|
|
|
+ let mut chunk_fd = File::create(&chunk_path).await?;
|
|
|
+ chunk_fd.write_all(chunk_slice).await?;
|
|
|
+
|
|
|
+ Ok(chunk_hash)
|
|
|
+ }
|
|
|
+
|
|
|
+ /// Fetch file metadata from Geode. Returns [`ChunkedFile`] which gives a list
|
|
|
+ /// of chunks and optionally file paths to the said chunks. Returns an error if
|
|
|
+ /// the read failed in any way (could also be the file does not exist).
|
|
|
+ pub async fn get(&self, file_hash: &blake3::Hash) -> Result<ChunkedFile> {
|
|
|
+ info!(target: "geode::get()", "[Geode] Getting file chunks for {}...", file_hash);
|
|
|
+ let mut file_path = self.files_path.clone();
|
|
|
+ file_path.push(file_hash.to_hex().as_str());
|
|
|
+
|
|
|
+ // Try to read the file metadata. If it's corrupt, return an error signalling
|
|
|
+ // that garbage collection needs to run.
|
|
|
+ let chunk_hashes = match Self::read_metadata(&file_path).await {
|
|
|
+ Ok(v) => v,
|
|
|
+ Err(e) => match e {
|
|
|
+ // If the file is not found, return according error.
|
|
|
+ Error::Io(err) if err == std::io::ErrorKind::NotFound => return Err(e),
|
|
|
+ // Anything else should tell the client to do garbage collection
|
|
|
+ _ => return Err(Error::GeodeNeedsGc),
|
|
|
+ },
|
|
|
+ };
|
|
|
+
|
|
|
+ let mut chunked_file = ChunkedFile::new(&chunk_hashes);
|
|
|
+
|
|
|
+ // Iterate over chunks and find which chunks we have available locally.
|
|
|
+ let mut buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ for (chunk_hash, chunk_path) in chunked_file.0.iter_mut() {
|
|
|
+ let mut c_path = self.chunks_path.clone();
|
|
|
+ c_path.push(chunk_hash.to_hex().as_str());
|
|
|
+
|
|
|
+ if !c_path.exists().await || !c_path.is_file().await {
|
|
|
+ // TODO: We should be aggressive here and remove the non-file.
|
|
|
+ continue
|
|
|
+ }
|
|
|
+
|
|
|
+ // Perform chunk consistency check
|
|
|
+ let mut chunk_fd = File::open(&c_path).await?;
|
|
|
+ let bytes_read = chunk_fd.read(&mut buf).await?;
|
|
|
+ let chunk_slice = &buf[..bytes_read];
|
|
|
+ let hashed_chunk = blake3::hash(chunk_slice);
|
|
|
+ if &hashed_chunk != chunk_hash {
|
|
|
+ // The chunk is corrupted/inconsistent. Garbage collection should run.
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ continue
|
|
|
+ }
|
|
|
+
|
|
|
+ *chunk_path = Some(c_path);
|
|
|
+ buf = [0u8; MAX_CHUNK_SIZE];
|
|
|
+ }
|
|
|
+
|
|
|
+ Ok(chunked_file)
|
|
|
+ }
|
|
|
+}
|